The only interesting thing I have to say here that hasn't been said is
that this test seems like a variation of Pair Programming, where one
person writes the tests and the other writes the code.  No better way
to ensure you only write code to make a failing test pass than to know
nothing about the problem space except the most recent test failure.

class Object
  def attribute(arg, &block)
    name = (arg.class == Hash ? arg.keys[0] : arg)
    define_method(name) do
      first_access_action(arg,&block) unless instance_eval("defined? "
+ "@" + name)
      instance_variable_get "@" + name
    end
    attr_writer name
    alias_method name+"?",name
  end

  def first_access_action(arg,&block)
    name = (arg.class == Hash ? arg.keys[0] : arg)
    send(name+"=",instance_eval(&block)) if block_given?
    send(name+"=",arg.values[0]) if arg.class == Hash
  end
end